def test_diag():
shape = (100,)
py_rand.set_seed()
py_vec = py_rand.randn(shape)
py_mat = py_matrix.diagonal_matrix(py_vec)
py_diag = py_matrix.diag(py_mat)
assert py_matrix.allclose(py_vec, py_diag), \
"python vec -> matrix -> vec failure: diag"
torch_vec = torch_rand.randn(shape)
torch_mat = torch_matrix.diagonal_matrix(torch_vec)
torch_diag = torch_matrix.diag(torch_mat)
assert torch_matrix.allclose(torch_vec, torch_diag), \
"torch vec -> matrix -> vec failure: diag"
def assert_close(pymat, torchmat, name, rtol=1e-05, atol=1e-06):
pytorchmat = torch_matrix.to_numpy_array(torchmat)
torchpymat = torch_matrix.float_tensor(pymat)
py_vs_torch = py_matrix.allclose(pymat, pytorchmat, rtol=rtol, atol=atol)
torch_vs_py = torch_matrix.allclose(torchmat, torchpymat, rtol=rtol, atol=atol)
if py_vs_torch and torch_vs_py:
return
if py_vs_torch and not torch_vs_py:
assert False,\
"{}: failure at torch allclose".format(name)
elif not py_vs_torch and torch_vs_py:
assert False, \
"{}: failure at python allclose".format(name)
else:
assert False, \
"{}: failure at both python and torch allclose".format(name)
def test_conversion():
shape = (100, 100)
py_rand.set_seed()
py_x = py_rand.rand(shape)
torch_x = torch_matrix.float_tensor(py_x)
py_torch_x = torch_matrix.to_numpy_array(torch_x)
assert py_matrix.allclose(py_x, py_torch_x), \
"python -> torch -> python failure"
torch_rand.set_seed()
torch_y = torch_rand.rand(shape)
py_y = torch_matrix.to_numpy_array(torch_y)
torch_py_y = torch_matrix.float_tensor(py_y)
assert torch_matrix.allclose(torch_y, torch_py_y), \
"torch -> python -> torch failure"
def test_fill_diagonal():
n = 10
py_mat = py_matrix.identity(n)
torch_mat = torch_matrix.identity(n)
fill_value = 2.0
py_mult = fill_value * py_mat
py_matrix.fill_diagonal(py_mat, fill_value)
assert py_matrix.allclose(py_mat, py_mult), \
"python fill != python multiplly for diagonal matrix"
torch_mult = fill_value * torch_mat
torch_matrix.fill_diagonal(torch_mat, fill_value)
assert torch_matrix.allclose(torch_mat, torch_mult), \
"torch fill != python multiplly for diagonal matrix"
assert_close(py_mat, torch_mat, "fill_diagonal")
def test_transpose():
shape = (100, 100)
py_rand.set_seed()
py_x = py_rand.rand(shape)
torch_x = torch_matrix.float_tensor(py_x)
py_x_T = py_matrix.transpose(py_x)
py_torch_x_T = torch_matrix.to_numpy_array(torch_matrix.transpose(torch_x))
assert py_matrix.allclose(py_x_T, py_torch_x_T), \
"python -> torch -> python failure: transpose"
torch_rand.set_seed()
torch_y = torch_rand.rand(shape)
py_y = torch_matrix.to_numpy_array(torch_y)
torch_y_T = torch_matrix.transpose(torch_y)
torch_py_y_T = torch_matrix.float_tensor(py_matrix.transpose(py_y))
assert torch_matrix.allclose(torch_y_T, torch_py_y_T), \
"torch -> python -> torch failure: transpose"